NVIDIA Unveils Nemotron 3 Ultra at GTC Taipei for Advanced AI Agents
NVIDIA launches Nemotron 3 Ultra, a powerful open model designed to power long-running AI agents at GTC Taipei. The new release promises significantly faster inference and reduced costs for complex enterprise tasks.
NVIDIA kicks off its GTC Taipei conference at COMPUTEX by introducing Nemotron 3 Ultra, an open model specifically built for long-running AI agents. Developed with contributions from the Nemotron Coalition, this 550-billion-parameter mixture-of-experts model handles complex tasks like architectural coding decisions, synthesizing large volumes of research, and verifying thousands of constraints. Industry leaders including Perplexity, Palantir, and ServiceNow are already adopting the technology to enhance their autonomous workflows.
Unlike traditional text generators, Nemotron 3 Ultra actively interprets information, plans steps, utilizes tools, and evaluates results across multiple turns to complete intricate enterprise tasks. The model delivers up to five times faster inference compared to previous versions and reduces the cost of complex agentic workloads by up to thirty percent. This efficiency allows AI systems to accomplish more work in less time, making autonomous agents more practical for widespread business use.
Major enterprise software companies are quickly integrating the new model into their existing platforms to build secure, long-running agents at scale. Aible is embedding Nemotron 3 Ultra into its AIbleClaw platform for diverse domain applications, while Glean is incorporating the model into its agent harness alongside a specialized agentic search model. Additional developers like Greptile are also leveraging the technology to push the boundaries of what autonomous AI agents achieve in software development and deep research.